Generative video tools promise a finished explainer for the price of a few credits. The credits turn out to be the cheapest part.
A client told us something recently that we haven’t stopped thinking about.
He’d been building an explainer video for his product with generative AI video tools. Every night. Every weekend. He’s a sharp guy who knows his product cold, and he attacked the problem the way smart people do: methodically, stubbornly, and with a heroic amount of coffee.
The first eight seconds were pretty good.
Everything after that fell apart. And after all those hours, he still didn’t have a video he’d be willing to put his name on.
That’s not a talent problem. He has an exceptional mind. It’s a math problem. So let’s do the math.
The tokens are the cheap part
Start with the number everyone checks first, because it’s honestly not that scary.
Google’s Gemini Omni Flash video model works out to about $0.10 per second of 720p video on the Gemini API. Runway’s Gen-4.5 model costs 12 credits per second, at a penny per credit on its API. Call it a dime or so for every second of footage.
Here’s the catch: you pay for every attempt, not every keeper. AI video is probabilistic. Same prompt, different result, every single time. When invideo documented a three-minute animated episode, the team generated 164 clips and kept 41, a 25% keep rate. That’s from a company that sells AI video tools. A 90-second short using a similar workflow took roughly 400 generations.
Four hundred eight-second generations at a dime a second comes to about $320. Real money, but not budget-breaking. If generation fees were the whole story, this would be a very short post.
They’re not the whole story.
Your nights and weekends have a price tag
The biggest cost of DIY AI video is the person doing it.
The median US marketing manager earns $166,790 a year, or about $80 an hour. The median software developer, often the subject-matter expert who knows the product best, earns $65.38 an hour. Those are wages alone, before benefits.
Now let’s be generous and say a DIY explainer takes 80 hours. That’s roughly a month of nights and weekends, which is about what our client described.
At a marketing manager’s wage, that’s around $6,400 of time. Add the generation fees and you’re past $6,700 before you’ve paid for music, voiceover, or editing software. For comparison, Wyzowl’s survey of production companies found a median quote of $5,400 for a 60-second animated explainer. (Its data dates from late 2022, so treat that as a ballpark, not gospel.)
And that $6,700 assumes the DIY version actually works. Our client’s attempts didn’t.
There’s also a cost that never shows up on an invoice: whatever those 80 hours would have built instead. The launch plan. The pipeline. The product. Sleep, even.
The eight-second wall
Remember how our client’s first eight seconds looked fine? That’s probably not a coincidence.
Google’s Veo 3.1 generates clips of up to eight seconds. You can extend a clip seven seconds at a time, but extensions are limited to 720p. So a 90-second explainer means a dozen or more separate generations that all need to look like they belong in the same movie.
That’s where things unravel. Each generation is a fresh guess. Faces drift between shots. Colors and lighting wander. Your product shows up looking like its own distant cousin. Even invideo’s guidance says a recurring character took about five generations just to lock its look, before any real shots began.
An explainer isn’t a pile of pretty clips. It’s an argument that builds for 90 seconds. Stitch it together from strangers, and the argument gets lost.
You can’t just fix the logo
In traditional production, a change request is surgical. Someone asks for a bigger logo, an editor makes the logo bigger, and everything else stays exactly where it was.
Generative video doesn’t do surgical. Change the prompt and you get a new shot, with a new set of surprises. Fix the logo, lose the hand. Fix the hand, and the background wanders off.
For tech companies, it gets worse. Explainers depend on accurate details: real UI, real feature names, and real workflows. Video models cheerfully invent buttons, settings screens, and text that looks like it was typed by an intoxicated raccoon. If your audience is IT pros or security buyers, a hallucinated admin console isn’t a quirk. It’s a reason not to trust you.
Credibility is the most expensive line item
Your explainer is often a buyer’s first real look at your product. How it’s made matters more than ever.
In research from The Harris Poll, the 4As, and Infillion, 73% of consumers said they’d be less likely to trust an ad they suspected was AI-generated, and 63% said they’d be less likely to buy from a brand using AI-generated ads. A Clutch survey of 601 consumers in August 2026 found 53% are less likely to buy from brands they know use AI in social content.
That Clutch survey also had a more nuanced finding: 42% view AI positively for educational content, and nearly 90% want brands to disclose AI-generated content. So audiences aren’t rejecting AI outright. They’re rejecting AI that feels sloppy or sneaky. A DIY explainer with drifting faces and a made-up interface manages both.
Who owns it, and will the tool still be there?
Two risks DIYers rarely price in.
Ownership. The US Copyright Office concluded in its January 2025 report that prompts alone don’t give users enough control to count as authors of the output. Footage generated purely from prompts may not be protectable, which means you might not be able to stop someone from reusing it. (We’re a video studio, not a law firm, so run this one past your lawyer.)
Platform risk. OpenAI shut down the Sora app on April 26, 2026, and its API on September 24. Google shut down its Veo 3.1 preview models in the Gemini API on October 22. If your entire workflow lives inside one tool, it can disappear halfway through your project.
Where DIY AI actually shines
Fair’s fair. We use AI at Honeycutt Inc. every day. Our team describes it as a bored intern who’s brilliant at the stupid stuff and has almost no taste. Useful. Never in charge.
DIY generative video is great for storyboards, animatics, pitch concepts, internal updates, and quick social posts. If the stakes are low and good enough really is good enough, go for it.
But your explainer isn’t a throwaway. It’s the 90 seconds a prospect uses to decide whether you’re worth another meeting.
The math, side by side
- Your team’s time. DIY: 80+ hours, about $6,400. Studio: two to three hours.
- Generation fees. DIY: about $320, often more. Studio: included.
- Studio fee. DIY: $0. Studio: a $5,400 median quote.
- Timeline. DIY: open-ended. Studio: two to three weeks at Honeycutt Inc.
- Edits. DIY: regenerate and hope. Studio: change what you asked for.
- Ownership. DIY: murky for prompt-only footage. Studio: human-made work.
- Finished video. DIY: not guaranteed. Studio: delivered.
Before you spend another weekend
Our client is smart, determined, and good at his job. That’s exactly why his time is too valuable to spend arguing with a prompt at midnight.
Already deep into a DIY project? Don’t toss it. We can often take what you’ve built and carry it across the last mile. We fix the continuity, rebuild the shots that won’t behave, swap in real UI, and finish it with professional voiceover and sound. Your weekends don’t have to be wasted. They just shouldn’t be the whole plan.
If you’re weighing DIY against a studio, book a call. We’ll tell you honestly whether AI, a studio, or a little of both makes sense for what you’re making. Sometimes the answer really is “do it yourself,” and we’ll say so.
